Telus Deploys Real-Time Accent Conversion on Call-Center Agents via Tomato.ai

Telus, through its Telus Digital unit, has deployed a live speech-to-speech system from Tomato.ai that modifies call-center agents' accents in real time. According to reporting by iPhone in Canada and The Globe and Mail, the tool targets what Telus calls "accent-related friction."

How It Works

The system processes call-center audio through a real-time speech-to-speech pipeline. While specific architectural details from Tomato.ai are not public, these systems typically combine automatic speech recognition (ASR), speaker and accent conversion models, and a neural vocoder. Latency-optimized inference is critical to avoid awkward pauses in conversation, and handling noisy call-center audio requires robust front-end ASR. The reported deployment focuses on offshore agents' voices.

Backlash and Industry Response

Labour groups have criticized the practice as deceptive and urged mandatory customer disclosure. Rogers and Bell, Canada's other major telecoms, told The Globe and Mail they have no plans to adopt similar technology. The public backlash has been swift.

Technical Considerations for Practitioners

For developers building contact-center AI pipelines, real-time voice conversion introduces operational tradeoffs: minimizing latency while preserving natural prosody, and maintaining robustness to background noise. Privacy compliance and worker consent are emerging regulatory risks. The Globe and Mail noted potential effects on voice-privacy regulations.

What to Watch

• Canadian regulatory guidance on disclosure requirements for voice-altering AI.
• Any technical disclosures from Tomato.ai about model latency and safeguards.
• Whether larger contact-center operators adopt transparency policies or technical audits.